Early detection and precise localization of insect pests are critical for the development of effective crop management strategies. To tackle such a challenge, a prior study proposed a Data Assimilation (DA) method that integrates a pheromone propagation model with data pheromone sensors to infer spatio-temporal pheromone emission maps and deduce insect localization. In continuity, this study proposes a Sequential Data Assimilation (SDA) method that iteratively refines sensor positioning by leveraging predicted pheromone plumes, enabling sensors to adaptively reposition toward areas of higher pheromone concentration.This approach combines data assimilation, large-scale physics modeling, and statistical optimization to improve localization accuracy over time. The method is evaluated through numerical experiments, including toy cases with unsteady wind conditions and multiple pheromone sources, as well as a realistic case based on real agricultural landscapes and meteorological data.Results demonstrate that the method significantly enhances the accuracy of pest localization compared to traditional data assimilation approaches. The sequential repositioning of sensors reduces errors in pheromone emission inference and reduces the false absence predictions from 100% to 0% within a few cycles, even in challenging scenarios such as unsteady wind or multiple emission sources.This study highlights the potential of SDA for robust pest localization, offering a promising tool for precision agriculture and sustainable pest management.
One third of the annual world's crop production is directly or indirectly damaged by insects, with an even increasing burden in a warming climate. Early detection of invasive insect pests is key for optimal treatment before infestation. Existing detection devices are based on pheromone traps: attracting pheromones are released to lure insects into the traps, with the number of captures indicating the population levels. Promising new sensors are on development to directly detect pheromones produced by the pests themselves and dispersed in the environment. Inferring the pheromone emission would allow locating the pest's habitat, before infestation. This early detection enables to perform pesticide-free elimination treatments and reduce the negative impact of agricultural practices on biodiversity, environment and human health, in a precision agriculture framework. In order to identify the sources of pheromone emission from signals produced by sensors spatially positioned in the landscape, the inference of the pheromone emission (inverse problem) is performed. In the present case, classical inference framework consists in combining the data from the pheromone sensors and the fluid mechanic-based pheromone concentration dispersion model that is a 2D reaction-diffusion-convection model. The proposed inference framework further incorporates into this combination additional a priori biological knowledge on pest behaviour (favourite habitat, insect clustering for reproduction, population dynamic behaviour...) [1]. This information is introduced to constrain the inference problem towards biologically relevant solutions. Different biology-informed constraints are tested, and the accuracy of the solutions of the inverse problems is assessed on simulated noisy data using a dedicated package [2]. In addition, optimal experimental design will be presented to deduce optimal sensor position in order to reduce the uncertainty of the inference and to improve the prediction of pest’s habitat localization.Reference:[1] Malou T., Parisey N., Adamczyk-Chauvat K., Vergu E., Laroche B., Calatayud P.-A., Lucas P. and Labarthe S. (2024). Biology-Informed inverse problems for insect pests detection using pheromone sensors. Submitted for publication. https://doi.org/10.5281/ZENODO.11506617[2] Malou T. and Labarthe S. (2024). Pherosensor-toolbox: a Python package for Biology-Informed Data Assimilation. Journal of Open Source Software, 29 (101), 6863. https://doi.org/10.21105/joss.06863.Acknowledgements:This work was carried out with the financial support of the French Research Agency through the Pherosensor project with grant agreement ANR-20-PCPA-0007.
Reconstructing gene regulatory networks from large-scale heterogeneous data is a key challenge in biology. In multi-omics data analysis, networks based on pairwise statistical association measures remain popular, as they are easy to build and understand. In the presence of mixed-type (discrete and continuous) data, however, the choice of good association measures remains an important issue. We propose here a novel approach based on the Gaussian copula, the parameters of which represent the links of the network. Novel properties of the model are obtained to guide the interpretation of the network. To estimate the copula parameters, we calculated a semiparametric pairwise likelihood for mixed data. In an extensive simulation study, we showed that the proposed estimation procedure was able to accurately estimate the copula correlation matrix. The proposed methodology was also applied to a real ICGC dataset on breast cancer, and is implemented in a freely available R package heterocop.
Classical joint modeling approaches often rely on competing risks or recurrent event formulations to describe complex processes involving evolving longitudinal biomarkers and discrete event occurrences, but these frameworks typically capture only limited aspects of the underlying event dynamics. We propose a general multi-state joint modeling framework that unifies longitudinal biomarker dynamics with multi-state time-to-event processes defined on arbitrary directed graphs. The proposed framework accommodates arbitrary directed transition graphs, nonlinear longitudinal submodels, and scalable inference via stochastic gradient descent. This formulation encompasses both Markovian and semi-Markovian transition structures, allowing recurrent cycles and terminal absorptions to be naturally represented. The longitudinal and event processes are linked through shared latent structures within nonlinear mixed-effects models, extending classical joint modeling formulations. We derive the complete likelihood, establish conditions for identifiability, and develop scalable inference procedures based on stochastic gradient descent to enable high-dimensional and large-scale applications. In addition, we formulate a dynamic prediction framework that provides individualized state-transition probabilities and personalized risk assessments along complex event trajectories. Through simulation and application to the PAQUID cohort, we demonstrate accurate parameter recovery and individualized prediction.
